Note: This dashboard contains the results of a predictive model. The author has tried to make it as accurate as possible. But the COVID-19 situation is changing quickly, and these models inevitably include some level of speculation.

Data

These are the countries included in the model:

'Albania,   Algeria,   Andorra,   Argentina,   Armenia,   Australia (total),   Austria,   Azerbaijan,   Bahrain,   Belgium,   Bosnia and Herzegovina,   Brazil,   Brunei,   Bulgaria,   Burkina Faso,   Canada (total),   Chile,   Colombia,   Costa Rica,   Croatia,   Cyprus,   Czechia,   Denmark,   Denmark (total),   Diamond Princess,   Dominican Republic,   Ecuador,   Egypt,   Estonia,   Finland,   France,   France (total),   Germany,   Ghana,   Greece,   Hong Kong,   Hungary,   Iceland,   India,   Indonesia,   Iran,   Iraq,   Ireland,   Israel,   Italy,   Japan,   Jordan,   Kazakhstan,   Korea, South,   Kuwait,   Latvia,   Lebanon,   Lithuania,   Luxembourg,   Malaysia,   Malta,   Mexico,   Moldova,   Morocco,   Netherlands,   New Zealand,   North Macedonia,   Norway,   Oman,   Pakistan,   Panama,   Peru,   Philippines,   Poland,   Portugal,   Qatar,   Romania,   Russia,   San Marino,   Saudi Arabia,   Senegal,   Serbia,   Singapore,   Slovakia,   Slovenia,   South Africa,   Spain,   Sri Lanka,   Sweden,   Switzerland,   Taiwan*,   Thailand,   Tunisia,   Turkey,   US,   Ukraine,   United Arab Emirates,   United Kingdom,   United Kingdom (total),   Uruguay,   Venezuela,   Vietnam'

Predicted Cases By Country

Select a country from the drop down list below to toggle the visualization.

Growth Rate Predictions 

Text(0.5, 0.98, 'Posterior of daily growth')

Model Diagnostics - Trace Plots

The following trace plots help to assess the convergence of the MCMC sampler. You can safely ignore this if not familiar with MCMC.

About This Analysis

This analysis was done by Thomas Wiecki. Interactive visualizations were created by Hamel Husain.

The model that we are building assumes exponential growth. This is definitely wrong because growth would just continue uninterrupted into the future. However, in the early phase of an epidemic it's a reasonable assumption.1

We assume a negative binomial likelihood as we are dealing with count data. A Poisson could also be used but the negative binomial allows us to also model the variance separately to give more flexibility.

The model is also hierarchical, pooling information from individual countries.

  1. This notebook gets up-to-date data from the "2019 Novel Coronavirus COVID-19 (2019-nCoV) Data Repository by Johns Hopkins CSSE" GitHub repository. This code is provided under the BSD-3 License. Link to original notebook.↩